Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Novel Convolution and Attention Mechanism-based Model for 6D Object Pose Estimation

Alexander Du, Xiujin Liu

TL;DR

PoseLecTr presents a graph-based encoder-decoder for 6D object pose estimation from monocular RGB images, addressing limitations of grid-based CNNs by explicitly modeling spatial relations with a graph and introducing Legendre convolution for numerical stability. The method integrates spatial-attention and self-attention distillation to refine feature selection, and uses a graph Laplacian framework to capture both local and global dependencies. On LINEMOD, Occluded LINEMOD, and YCB-Video, PoseLecTr demonstrates state-competitive performance, with ablations showing the Legendre kernel and attention components as key contributors to robustness in cluttered and occluded scenes. The work advances practical 6D pose estimation by combining a graph-based representation with mathematically stable convolution and targeted attention mechanisms, improving accuracy for texture-less, symmetric, and occluded objects in realistic settings.

Abstract

This paper proposes PoseLecTr, a graph-based encoder-decoder framework that integrates a novel Legendre convolution with attention mechanisms for six-degree-of-freedom (6-DOF) object pose estimation from monocular RGB images. Conventional learning-based approaches predominantly rely on grid-structured convolutions, which can limit their ability to model higher-order and long-range dependencies among image features, especially in cluttered or occluded scenes. PoseLecTr addresses this limitation by constructing a graph representation from image features, where spatial relationships are explicitly modeled through graph connectivity. The proposed framework incorporates a Legendre convolution layer to improve numerical stability in graph convolution, together with spatial-attention and self-attention distillation to enhance feature selection. Experiments conducted on the LINEMOD, Occluded LINEMOD, and YCB-VIDEO datasets demonstrate that our method achieves competitive performance and shows consistent improvements across a wide range of objects and scene complexities.

A Novel Convolution and Attention Mechanism-based Model for 6D Object Pose Estimation

TL;DR

PoseLecTr presents a graph-based encoder-decoder for 6D object pose estimation from monocular RGB images, addressing limitations of grid-based CNNs by explicitly modeling spatial relations with a graph and introducing Legendre convolution for numerical stability. The method integrates spatial-attention and self-attention distillation to refine feature selection, and uses a graph Laplacian framework to capture both local and global dependencies. On LINEMOD, Occluded LINEMOD, and YCB-Video, PoseLecTr demonstrates state-competitive performance, with ablations showing the Legendre kernel and attention components as key contributors to robustness in cluttered and occluded scenes. The work advances practical 6D pose estimation by combining a graph-based representation with mathematically stable convolution and targeted attention mechanisms, improving accuracy for texture-less, symmetric, and occluded objects in realistic settings.

Abstract

This paper proposes PoseLecTr, a graph-based encoder-decoder framework that integrates a novel Legendre convolution with attention mechanisms for six-degree-of-freedom (6-DOF) object pose estimation from monocular RGB images. Conventional learning-based approaches predominantly rely on grid-structured convolutions, which can limit their ability to model higher-order and long-range dependencies among image features, especially in cluttered or occluded scenes. PoseLecTr addresses this limitation by constructing a graph representation from image features, where spatial relationships are explicitly modeled through graph connectivity. The proposed framework incorporates a Legendre convolution layer to improve numerical stability in graph convolution, together with spatial-attention and self-attention distillation to enhance feature selection. Experiments conducted on the LINEMOD, Occluded LINEMOD, and YCB-VIDEO datasets demonstrate that our method achieves competitive performance and shows consistent improvements across a wide range of objects and scene complexities.
Paper Structure (29 sections, 11 equations, 2 figures, 3 tables)

This paper contains 29 sections, 11 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Graph-Based Representation
  3. Legendre Convolution
  4. Spatial-Attention and Self-Attention Distillation
  5. Related Works
  6. Methodology
  7. Architecture
  8. Input and Output
  9. Graph Neural Network
  10. Legendre Convolution Layer
  11. Orthogonality
  12. Numerical Stability
  13. Recursion
  14. Spatial Feature Self-Attention
  15. Loss Function
  16. ...and 14 more sections

Figures (2)

  • Figure 1: The architecture of PoseLecTr
  • Figure 2: Visualization of Ablation Studies